2D fatigue crack propagation in rails taking into account actual

plastic stresses
B. Troll
a,b,
, M.-C. Baietto
a
, A. Gravouil
a,c
, S.H. Mai
b
, B. Prabel
d
a
Universit de Lyon, CNRS INSA-Lyon, LaMCoS, UMR5259, 20 Avenue Albert Einstein, F69621 Villeurbanne Cedex, France
b
SNCF, Direction Innovation et Recherche, 40 Avenue des Terroirs de France, 75611 Paris Cedex 12, France
c
Institut Universitaire de France, France
d
CEA, DEN, DM2S, SEMT, DYN, CEA-Saclay, 91191 Gif-sur-Yvette Cedex, France
a r t i c l e i n f o
Article history:
Available online xxxx
Keywords:
Rolling contact fatigue
Frictional cracks
Residual stresses
Crack propagation
X-FEM
a b s t r a c t
Due to increase axle loads, repeated passages of the wheels, rolling contact fatigue cracks
initiate in the surface or subsurface of the rails. These defects can propagate and lead to the
rail failure. A two-scale frictional contact fatigue crack model developed within the X-FEM
framework is used to address the cracked rail problem, calculate the stress intensity fac-
tors, perform crack propagation and fatigue life prediction. Actual plastic stresses, quanti-
ed through a dedicated software developed by SNCF within the consortium IDR2, are
taken into account in the propagation simulation via projection of the asymptotic mechan-
ical elds. The effects of those actual plastic stresses are investigated on the crack growth
path and rate. Interactions between multiple cracks of a network are also analyzed.
2014 Elsevier Ltd. All rights reserved.
1. Introduction
The repeated train trafc leads to crack initiation at the rail surface or within the subsurface, up to the development of
very complex 3D crack network, like squats and head-checks. Rail failure may occur and potentially derailment too. Costly
maintenance operations are deployed to avoid such a situation.
Rolling contact loading is a time-dependent, multi-axial and nonproportional loading. This cyclic loading leads to an
asymptotic stress state within the rail that inuences the crack behavior. The cracks are also submitted to large compressive
stresses inducing crack face closure and frictional contact.
Previous works about fatigue crack growth in the rails are available in the literature. The role of liquid entrapment using
nite element analysis [13] or boundary element method [4,5] was analyzed on the crack growth mechanism. The inuence
of different parameters, such as elastic foundation [3,6], the initial crack geometry or the crack face friction coefcient
[3,5,7,8] was also investigated on the stress intensity factors (SIFs) or on the crack propagation path [9]. It is well known that
the distribution of the residual stress eld affects the crack propagation behavior [1015], but most of previous studies use
simplied techniques. This residual stress eld is accounted as the addition of uniform tensile or compressive conditions at
the boundary of the considered domain [8,16,17].
To improve the understanding of the crack initiation and propagation mechanisms, a global methodology has been devel-
oped thanks to a long-term collaboration between French railway organizations (SNCF, RATP, RFF), rail producer (Tata Steel)
http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
0013-7944/ 2014 Elsevier Ltd. All rights reserved.

Corresponding author at: Universit de Lyon, CNRS INSA-Lyon, LaMCoS, UMR5259, 20 Avenue Albert Einstein, F69621 Villeurbanne Cedex, France. Tel.:
+33 171323324.
E-mail address: benoit.trolle@sncf.fr (B. Troll).
Engineering Fracture Mechanics xxx (2014) xxxxxx
Contents lists available at ScienceDirect
Engineering Fracture Mechanics
j our nal homepage: www. el sevi er. com/ l ocat e/ engf r acmech
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
and research institutes and universities (LaMCoS/INSA Lyon, LMS/Polytechnique, MECAMIX, IFSTTAR) within the consortium
IDR2 (Initiative for Development and for Research on Rails) based on a theoretical, numerical and experimental approach.
Concerning the numerical simulation, it starts from the train trafc to a fatigue life assessment [18].
v Step 1: a railway multi-body dynamics is performed, accounting for the actual rail and wheel geometries and load con-
ditions, to determine contact conditions at the wheel-rail interface.
v Step 2: the stabilized cyclic mechanical state of the rail is calculated using a 3D elasto-plastic nite element simulation
and an original, timecost efcient direct stationary algorithm.
v Step 3: a fatigue analysis of the rail is performed according to Dang Van criterion to dene the critical sites for crack ini-
tiation within the rail.
v Step 4: the crack (s) are considered and their behavior and growth are analyzed.
This paper focuses on this latter point. A two dimensional Extended Finite element model accounted for actual residual
stresses and frictional contact between crack faces is combined with a mixed mode nonproportional crack propagation
model previously developed. In Section 2, the theoretical basis of the two-scale XFEM strategy for cracked body problem
with frictional contact at the interface accounting for actual residual stresses is presented. Then, in Section 3 the fatigue crack
growth procedure is briey introduced. Section 4 illustrates the reliability of the results. Parametric studies are then per-
formed to analyze the inuence of the tangential loading, the friction coefcient between the crack faces and the initial crack
orientation on the crack growth path and rate. In the last part of this section, the actual residual stress inuence is studied
and the role of neighboring cracks in the crack propagation mechanism is emphasized.
2. Two-scale XFEM strategy for cracked body problem with frictional contact at the interface accounting for actual
residual stresses
2.1. Two-scale strategy for cracked body problem with frictional contact at the interface
Different authors have proposed mixed weak formulation to take into account the contact and the friction between the
crack faces (Fig. 1) within the X-FEM framework [1922]. We here use the approach proposed in [20] and consider a cracked
body X R
3
. Contact and friction can occur along the crack faces C

and C

(Fig. 2(a)). Under small displacement and small

strain assumptions, the interface C (C = C

) is assumed as an autonomous entity with its own behavior possibly

Nomenclature
(0) initial elastic strain for the cracked structure problem

p
initial plastic strain for the cracked structure problem
r(0) initial elastic stresses for the cracked structure problem
r
e
elastic stress tensor
r
plastic
initial plastic stresses for the cracked structure problem

e
elastic strain tensor
da
dN
crack growth rate
l
crack
friction coefcient between the crack faces
l
wheelrail
wheel-rail friction coeffcient
r Cauchy stress tensor
K hook tensor
t local traction eld
u global displacement eld in the structure
w local displacement eld
h initial angle between the crack and the running surface
a Hertzian contact semi-length
da crack extension length
dN cycle jump
k
1
mode I stress intensity factor at the tip of a virtual crack extension
k
2
mode II stress intensity factor at the tip of a virtual crack extension
K
eq
equivalent stress intensity factor at the actual crack tip
K
II
mode II stress intensity factor at the actual crack tip
K
I
mode I stress intensity factor at the actual crack tip
l initial crack length
P
max
maximal Hertzian pressure
x
c
distance between the crack mouth and the maximal Hertzian pressure
2 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
nonlinear. This fracture problem is divided into a global (X C in Fig. 2(b)) and a local (crack scale in Fig. 2(c)) problem. The
global problem is dened with its own primal and dual variables, u the displacement eld and r the Cauchy stress tensor
respectively. The local problem is dened with its own primal and dual variables, w is the interface displacement eld
and t is the interface traction eld respectively. Let n be the outward unit normal to X C and n
C
is the outward unit normal
to C.
We assume quasi-static formulation and write the governing equations as follows for the global problem at a given time
t [0; T[:
div r f = 0 in X C
c
(1)
rn = s on @
2
X (2)
u = v on @
1
X (3)
where s and v are the prescribed load on @
2
X and displacement on @
1
X respectively, and f the body force vector. Both global
and local problems obey a constitutive law possibly nonlinear in the bulk and on the interface respectively:
r = r(u; u
i
; t) in X C
c
(4)
t = t(w; w
i
; t) on C
c
(5)
where u
i
and w
i
are the internal variables of the global and local problem respectively and t a given time. Furthermore, we
add the coupling between the global and the local problems both on the primal and dual variables:
u = w on C
c
(6)
rn = t on C
c
(7)
Eqs. (((1)(3), (6), (7)) consist in the strong formulation of the crack model with contact and friction between the crack faces.
As already shown in [23], an equivalent three eld weak formulation (the global displacement eld u, the local displacement
eld w and the local tractions eld t) at a given time t [0; T[ can be introduced (see [20,24]).
Fig. 1. Schematic representation of the phenomena occurring at the crack scale.
(a)
(b)
(c)
Fig. 2. Cracked body problem (a), divided into a global structure problem (b) and a local interface problem (c).
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 3
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
0 =
_
X
r : (u
+
)dX
_
X
f u
+
dX
_
@
2
X
s u
+
dS
_
Cc
t u
+
dS
_
Cc
(p t) w
+
dS

_
Cc
(u w) t
+
dS \u
+
U
0
; \w
+
W; \t
+
K; \t [0; T[ (8)
with
u U; U = u H
1
(X C
c
)=u = v on @
1
X (9)
u
+
U
0
; U
0
= u H
1
(X C
c
)=u = 0 on @
1
X (10)
w W; w
+
W; W = w H
1
(C
c
) (11)
t K; t
+
K; K = t L
2
(C
c
) (12)
and is equivalent to the previous strong formulation (1, 2,3, 6,7) (see [23]). This weak formulation (8) allows a globallocal X-
FEM formulation handling successfully the structure scale (global), the crack and the interfacial nonlinear local scales. Here,
an homogeneous, isotropic material with an elastic linear behavior is considered. The stress strain law is then written
according to the fourth order Hooke tensor K:
r = K (u) in X C
c
(13)
Furthermore, at the local scale, interface quantities (w; t) obey contact unilateral law and Coulombs friction law for the
normal and tangential problem respectively. w and t interface elds on the crack faces C

c
and C

c
are expressed in the local
basis (n
c
; t
c
) relative to the crack
w = w
N
n
c
w
T
t
c
and t = t
N
n
c
t
T
t
c
(14)
where (w
N
; t
N
) and (w
T
; t
T
) are normal and tangential scalar interface elds, respectively. Further relative interface displace-
ments (opening and sliding) between the crack faces at a given interface point x are given by:
[w
N
(x; t)[ = w

N
(x; t) w

N
(x; t) and [w
T
(x; t)[ = w

T
(x; t) w

T
(x; t) (15)
This yields for the local equations of the interface behavior:
v opening [w
N
(x; t)[ > 0 t

(x; t) = t

(x; t) = 0
v contact [w
N
(x; t)[ = 0 t

(x; t) = t

(x; t)
v sticking [t
T
(x; t)[ < l
c
[t
N
(x; t)[ D[w
T
(x; t)[ = 0
v sliding [t
T
(x; t)[ = l
c
[t
N
(x; t)[ c > 0=D[w
T
(x; t)[ = ct

T
(x; t)
(16)
where [w
N
(x; t)[ and [w
T
(x; t)[ are the normal and tangential jump at the interface, l
c
is the friction coefcient at the interface
2.2. X-FEM discretization and nonlinear solver dedicated to interfacial frictional contact
The eXtended Finite Element Method [25] enriches the approximation space so that it allows to account for discontinu-
ities like cracks. It addresses nonlinear problems, no explicit crack mesh and no remeshing during the crack growth process.
Compared to the FEM, X-FEM consists in introducing additional functions to the discretization basis following the partition
of unity framework [26]. In this respect, due to the presence of a crack, one denes an appropriate approximation displace-
ment eld composed of three terms: (i) the rst term or the standard nite element approximation, characterizing the con-
tinuum behavior of the body, (ii) the second term corresponding to an arbitrary displacement discontinuity along the crack
faces, (iii) and the latter capturing the asymptotic behavior of the stresses near the crack tip. One assumes that the state vec-
tor X
n
= (u
n
; r
n
; w
n
; t
n
) is known at time t
n
. Within a quasi-static incremental framework, the next stage consists of calcu-
lating the unknown state vector X
n1
at time step t
n1
. The LATIN nonlinear solver consists of an iterative strategy
between a local stage from X
i
n
to X
i
1
2
n
and a global stage from X
i
1
2
n
to X
i1
n
. The local stage corresponds to a set of local equa-
tions, possibly nonlinear (Eq. 16), and the global stage to a set of global linear equations (Eq. 8). This two-steps approach
requires search directions E
GL
and E
LG
between the set of equations (L) and (G). The process is repeated until convergence
is reached.
The introduction of the X-FEM discretization of the elds and the conditions introduced by the stabilization process [24]
lead to the discretized linear system (25). A priori formula were recently proposed to set up the parameters controlling the
convergence rate (search direction and stabilization parameter) to obtain the best convergence rate possible [27] and
improve the method efciency by reducing the computing time.
K 0 K
ut
0 K
ww
K
wt
K
T
ut
K
T
wt
K
tt
_

_
_

_
U
i1
W
i1
T
i1
_
_
_
_
_
_ =
F
K
wt
T
i
1
2
K
ww
W
i
1
2
K
tt
T
i
1
2
_
_
_
_
_
_ (17)
4 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
The evaluation of the mechanical state in the rail is crucial for the prediction of its durability. The repeated contact stres-
ses at the wheel/rail interface during rolling motion are responsible for plastic deformations within the contact zone. The
residual stress eld in the rail section were studied in a lot of works [1015,28,29]. Nevertheless when dealing with crack
propagation simulation, residual stresses are basically introduced in the model applying forces [8,16,17] at the boundaries
of the considered domain. The current study focuses on the analysis of the inuence of actual computed residual stresses on
crack propagation.
2.3. Introduction of actual plastic stresses in the X-FEM strategy
The residual stresses considered here correspond to the plastic strain accumulation due to the repeated passages of the
wheels on the rails and to the manufacturing process [10]. They result from the rail cyclic stress and strain computation
under the action of a great number of these contact loads. The difculty lays on the fact that plastic deformations and high
stress gradients can occur in the area near the rail-wheel contact zone due to the high level of the rolling-sliding contact
loads. Therefore, very rened nite element meshes are used to capture accurately the stress gradients. Simulating a great
number of passages by running classical nite element simulations based on classical integration schemes would be very
lengthy and require considerable computing ressources. To avoid these drawbacks, alternative methods were proposed like
the direct stationary method [2931], going straight to the steady state in the case of a repeatedly moving load and applied
to rail modeling. It is based on the following ideas:
v the simulation is performed in a reference moving with the load, i.e., the wheel reference, that allows to take
advantage of the time independance of mechanical quantities;
v the regime is supposed to be stationary in this reference;
v the rail material ow is modeled using Eulerian formulations;
v elastoplastic behavior such as elastic or plastic shakedown can be considered;
v rened meshing is only necessary in the vicinity of the contact zone;
This alternative method allows proposing a very time cost-efcient computing algorithm and easy to use. An elastoplastic
constitutive law with a linear kinematic hardening is used for the steel of the rail.
Using this algorithm, two kinds of asymptotic states may be obtained. The former is elastic shakedown (Fig. 3(a)) where
the limit cycle is elastic or plastic shakedown (Fig. 3(b)) with an elasto-plastic limit cycle. The latter is the ratcheting for
which plastic strains increase continously without reaching an asymptotic state and thus a limit cycle. The attention is
focused here on elastic shakedown. Plastic strains are plotted in Fig. 4(a)(c) for elastic and plastic shakedown respectively.
Elastic shakedown is characterized by all components of the plastic deformation tensor constant along all the streamlines,
here the horizontal lines of the mesh whereas it is not the case for plastic shakedown.
The problem is expressed as follows. In the initial state (t = ), the rail is considered as a virgin structure. At t = 0, it is
assumed that the rail is in the asymptotic state and that shakedown has occured before crack initiation. We then perform
fatigue crack growth simulation (see Fig. 5).
For a given time t, the equilibrium equations, the constitutive law and the boundary equations are summarized hereafter:
Equilibrium : div r
e
(t) = 0 in X
Constitutive law : r
e
(t) = K
e
(u(t)) in X
Dirichlet condition : r
e
(t)n = F(t) on @
2
X
Neumann condition : u(t) = u
d
(t) on @
1
X
(18)
Fig. 3. Rail elastoplastic behavior: asymptotic state.
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 5
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
with
r
e
(t) = r(0) Dr(t) (19)
(t) = (0) D(t) (20)
where r(0) are the elastic stresses related to (0) (Fig. 6). If no loading is assumed, we only have plastic strains
p
and the
residual stresses are the plastic stesses dened by r
plastic
= K
p
Moreover
(t) =
p

e
(t) (21)
Multiplying (21) by K the left and right hand sides and introducing (20) in the left hand side, we obtain:
K((0) D(t)) = K
p
r(t) (22)
Fig. 4.
xx
plastic under loading in the case of elastic shakedown (b) and plastic shakedown (c) in the area of interest (a).
Fig. 5. Notation for the introduction of residual stresses.
Fig. 6. Output of the stationary algorithm.
6 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
We now introduce (22) into (19) to express Dr(t) as:
Dr(t) = K(D(t)
P
) K(0) r(0) (23)
which is equivalent to:
KD(t) Dr(t) = K
p
K(0) r(0) ( ) (24)
This equation is introduced into the Eq. (8). The linear system (Eq. (25)) is obtained with DU
i1
instead of U
i1
as unknown
vector. Before post-processing the results to compute the stress intensity factors, an update step should be performed in the
same way than in Eqs. (19) and (20) to obtain the eld U.
K 0 K
ut
0 K
ww
K
wt
K
T
ut
K
T
wt
K
tt
_

_
_

_
DU
i1
W
i1
T
i1
_
_
_
_
_
_ =
F F(0)
K
wt
T
i
1
2
K
ww
W
i
1
2
K
tt
T
i
1
2
_
_
_
_
_
_ (25)
with:
F(0) =< F
l
(0) > (26)
F
l
(0) =
_
X
Tr K
p
(K(0) r(0))
_ _
(/
l
)
_
dX (27)
The internal generalized force F
l
(0) has to be added to the right hand side of the linear system. Its estimation is based on
elds computed on the stationary method mesh. The method to compute the residual stresses assumes no gradient in the
moving load direction. The mesh is hence designed by a domain translation according to this direction. Different require-
ments are needed for the mesh created for the crack propagation simulation. Indeed, a ne mesh under the moving
wheel-rail contact is needed to capture the stress and strain gradients, but an even ner mesh in the crack propagation area
is necessary to determine accurately the K-dominance zone and the stress intensity factors at the crack tip. To reduce the
number of degrees of freedom and optimize the computing ressources, coarse and ne mesh domains are designed versus
either the residual stress computation or the crack propagation simulation. Therefore residual elds estimated according to
the stationary method with a coarser mesh are transferred onto the ner one dened for the propagation.
2.4. Projection step to introduce the generalized forces on the discretization used for the crack propagation simulation
The third step is concerned with the projection of those residual elds onto the mesh specially designed for the crack
propagation. Two main categories of methods exist to project mechanical elds between meshes with different renement.
Very briey, either the elds are projected from mesh 1 to mesh 2 and a balance step is introduced to ensure the respect of
the equilibrium on mesh 2 or only part of the elds are projected on mesh 2, computed according to the constitutive law and
the local variables are updated. This is this method which is used in this work. Under the assumptions of elastic shakedown
and considering an unloaded body we therefore have u(0) = (0) = r(0) = 0 and we only consider the plastic stresses:
F
l
(0) =
_
X
Tr K
p
_ _
/
l
( )
_
dX (28)
The plastic stresses are dened by K
p
= r
plastic
and
DU
i1
W
i1
T
i1
_
_
_
_
_
_ =
U
i1
W
i1
T
i1
_
_
_
_
_
_ (29)
The plastic stress eld projection onto the ner mesh followed by the computation of the equivalent generalized forces on
this mesh allows to enforce the equilibriumof the structure afterwards. Initially, the eld r
plastic
is dened at the Gauss points
of the coarse discretization. The projection is performed in two steps. The element eld is rst computed on the nodes of the
coarse mesh elements using an interpolation based on the shape functions and a least square to compute the value at a given
node. Then the projection is performed element by element from the coarse mesh m
1
to the ne mesh m
2
using the linear
shape functions as interpolation functions. The interpolations could have been performed using different functions than the
shape functions. This choice has been done for practical reasons. The last step consists in computing the generalized forces
corresponding to the projected plastic stresses on all the nodes of the mesh m
2
(Fig. 7).
As the material is assumed to behave elastically when dealing with the crack propagation simulation, an initial elastic
resolution is performed accounting only for generalized forces corresponding to the projected plastic stresses. The constitu-
tive law is automatically veried and the stress elds are balanced on the ne discretization. The asymptotic mechanical
elds are then projected on the ne mesh designed for crack propagation analysis (see Fig. 8). Those elds are the initial state
of the propagation simulation. This state is considered as permanent and nonuniform. No redistribution of plastic stresses is
performed throughout the crack growth.
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 7
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
3. Fatigue crack growth procedure
Rolling fatigue cracks experience nonproportional mixed mode loading during the load traveling corresponding to a char-
acteristic sequence of tensile stresses followed by a shear cycle, with complex sequences of opening/contact and stick/slip
distributions along the crack faces. These distributions govern the stress intensity factors and the growth behavior, both rate
and path. Some studies [32] on nonproportional loading have shown the inability of classical criteria like the Maximum Tan-
gential Stress MTS, the Maximum Shear Stress MSS, the minimum strain energy density S
min
, the maximum energy release
rate G
max
, etc. to determine the crack propagation direction under out of phase loading histories. The crack growth path h
direction and rate da=dN are here predicted according to Hourlier and Pineaus criterion [33]. This criterion was successfully
used previously under both rolling [8] and fretting fatigue [3436]. The crack path predictions were in very good agreement
with experimental observations.
The mixed-mode fatigue crack propagation model is briey recalled here. A more detailed explanation can be found in [8]
and more recently in [36]. The rst step consists in computing the stress intensity factors (SIFs) histories at crack tip during
the load traveling over the rail (one wheel passage) dened as one load cycle. This load cycle is divided into time steps cor-
responding to successive wheel positions with respect to the crack location. The crack body problem is solved at each time
step using the two scale strategy (2.1). The SIFs are computed using integral methods.
The second step is concerned with the crack growth path and rate determination. The crack growth path is predicted
according to Hourlier and Pineaus criterion [36] It assumes that a crack follows the easiest path available that is the path
along the direction b that maximizes the growth rate at the tip of an innitesimal virtual part s of the kinked crack. Values
of k
1
and k
2
, the local stress intensity factors at the tip of the virtual innitesimal part s inclined at an angle b of the original
Fig. 7. Projection procedure of the plastic stresses from a mesh m
1
to a mesh m
2
.
Fig. 8. 2D projection of the asymptotic mechanical elds computed on mesh m
1
on the mesh used for the crack propagation simulation (mesh m
2
).
8 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
crack, are calculated from K
I
and K
II
versus the angle b (Fig. 9). Corresponding Dk
1
and Dk
2
range are then calculated for each
angle b. A dedicated mixed-mode propagation law is then used to predict the crack growth rate [8], based on an equivalent
stress intensity factor range.
da
dN
= 2:10
9
(DK
eq
)
3:33
(30)
with
DK
eq
= DK
2
I
0:772 + DK
2
II
_ _
0:5
(31)
A crack extension is added to the current crack and the corresponding increment of number of cycles is estimated according
to the crack propagation law. This procedure is repeated until either a critical crack size or a given number of cycles is
reached (see Fig. 10).
4. Validation: X-FEM modeling of a surface breaking crack under a moving Hertzian loading
In this section, results obtained according to the X-FEM model previously described here are compared with results from
the litterature. The inuence of different factors on the SIFs KI and KII and the crack growth behavior are analyzed. Factors
like the wheel-rail friction coefcient, the crack face friction coefcient, the crack inclination with respect to the free surface,
the plastic stresses and the crack network interactions are investigated.
Sets of wheel/rail contact forces, wheel diameters and crack morphologies used here for the comparison are those dened
in [8]. The fatigue crack model is based on the crack analogy with continuous distributions of dislocations combined with
unilateral contact analysis. Detailed calculations of mode I and mode II effective SIFs ranges under nonproportional sequen-
tial loading cycles can be performed at crack tips, for any crack geometry (straight or kinked), crack number, accounting for
interactions between cracks [37,38]. A 6 mm long crack inclined at 15 with respect to a half-plane (the rail) is considered.
Fig. 9. Parameters introduced in the Hourlier and Pineaus criterion to compute the propagation angle b.
Fig. 10. Flowchart of the propagation procedure.
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 9
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
The crack friction coefcient is assumed to be equal to 0:1. A fully sliding Hertzian contact load is considered with peak pres-
sure, contact length and friction coefcient equal respectively to P
max
= 845 MPa, 2a = 13:5 mm, l
wheelrail
= 0:4 (see Fig. 11).
A rectangular domain, with dimensions L = 300 mm times H = 172 mm, is considered as a half-plane. The nite element
mesh used is presented in the Fig. 12. The rened zone is a rectangular domain containing the crack which size is automat-
ically adapted. For the reference case and in the initial conguration, the dimensions are l = 10 mm and h = 4 mm. The ratio
between the biggest and the smallest elements is higher than 500. The ne mesh domain is obviously located under the con-
tact zone to capture accurately the stress and strain gradients. Further, in the area where the crack will propagate, the mesh
is even ner to ensure a reliable modeling of the K-dominance zone. The SIF obtained according to both models are presented
Fig. 13 versus x
c
=a. The leading edge of the load distribution is located at the surface breaking tip of the crack at x
c
=a = 1.
Note that in [8] the SIF computation is based on the displacement jump between the crack faces whereas an energetic
approach and domain integral at the crack tip vicinity are used here. Therefore, different SIFs are obtained according to both
Fig. 11. Denition of the parameters.
Fig. 12. Multi-scale mesh to capture the different scales of the problem.
30
40
50
-20
-10
0
10
20
-6,0 -4,0 -2,0 0,0 2,0 4,0 6,0
K
I

(
M
P
a
.
m




)
x
c
/a
0
.
5
30
40
50
-20
-10
0
10
20
-6,0 -4,0 -2,0 0,0 2,0 4,0 6,0
K
I
I

(
M
P
a
.
m




)
x
c
/a
0
.
5
KI [Dubourg2002]
KI CAST3M
KII [Dubourg2002]
KII CAST3M
KII [Dubourg2002]
KII CAST3M
Fig. 13. Comparison of (a) K
I
and (b) K
II
values obtained using CASTEM and [8].
10 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
methods in the particular case of a crack under full sticking conditions during several steps of the moving load. Constant K
II
values are obtained according to [8] whereas slightly varying ones are obtained here due the stress variations during the load
traveling. This effect can be observed in the Fig. 13 for the maximun K
II
values using our model (CAST3M [39] in Fig. 13).
Despite those differences the results are in very good agreement.
5. Parametric study
The inuence of different parameters on crack growth path and rate are investigated. Parameters corresponding to the so-
called reference case are listed in Table 1.
5.1. Inuence of the friction coefcient between the wheel and the rail
The crack inclination is imposed as well as the direction of the moving load, from negative to positive x values. The inu-
ence of the friction coefcient between the wheel and the rail is analyzed and the effects of both positive and negative values
are considered. The convention used is dened in Fig. 11 as well as the direction of the moving load. Values of 0.025, 0.1 and
0.4 are considered. For x
c
=a < 1, the load distribution has not moved over the crack whereas for x
c
=a > 1, the load is located
over the crack before traveling away. The tangential load and the normal load effects on the relative sliding and opening
along the crack faces can either be added or substracted depending on the sign of the friction coefcient and on the position
of the load with respect to the crack. This is illustrated in Figs. 14(b) and 15(b) for K
II
and Figs. 14(a) and 15(a) for K
I
.
The numerical crack propagation is then analyzed. The procedure described in Fig. 10 is used. The crack path is deter-
mined at each numerical cycle, the crack growth is simulated with a 4% crack total length increment and the corresponding
increment of number of cycles is quantied. 25 numerical load cycles are simulated, except if a crack self-arrests or branches
upwards out of the domain. For the crack propagation, a higher friction coefcient between the wheel and the rail causes an
earlier branching. DK
II
is increased and the crack branches. Depending on the orientation of the tangential load, the cracks
branch earlier upwards to the rail surface (Fig. 16(a)) or downwards in the rail web (Fig. 16(b)).
5.2. Inuence of the crack friction coefcient
The inuence of the friction coefcient between the crack faces is studied, all the other parameters being unchanged (see
Table 1). An increasing in the crack friction coefcient decreases the relative sliding between the crack faces and conse-
quently reduces DK
II
(see Fig. 17(b)) whereas it has obviously no inuence on the crack opening and on DK
I
(see
Fig. 17(a)). The crack friction coefcient and the wheel-rail friction coefcient have thus opposite inuences on the crack
growth rate [8,17].
Table 1
Reference parameters for the parametric study.
l (mm) h () l
crack
l
wheelrail
P
max
(MPa) 2a (mm)
6 15 0.5 0.025 845 13.5
0.025
0.1
0.4
0.025
0.1
0.4
6.00 4.00 2.00 0.00 2.00 4.00 6.00
6.00 4.00 2.00 0.00 2.00 4.00 6.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
25.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
25.00
Fig. 14. (a) K
I
and (b) K
II
computed for different positive friction coefcients l
wheelrail
> 0 (braking situation) (l = 6 mm, h = 15, l
crack
= 0:5).
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 11
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
-0.025
-0.1
-0.4
-0.025
-0.1
-0.4
6.00 4.00 2.00 0.00 2.00 4.00 6.00
6.00 4.00 2.00 0.00 2.00 4.00 6.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
25.00
20.00
25.00
30.00
-35.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
25.00
20.00
25.00
30.00
-35.00
Fig. 15. (a) K
I
and (b) K
II
computed for different negative friction coefcients l
wheelrail
< 0 (speeding up condition) (l = 6 mm, h = 15, l
crack
= 0:5).
Fig. 16. Crack propagation path for different friction coefcient between the wheel and the rail in braking situation (a) and speeding up condition (b).
0.01
0.1
0.5
0.01
0.1
0.5
6.00 4.00 2.00 0.00 2.00 4.00 6.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
6.00 4.00 2.00 0.00 2.00 4.00 6.00
Fig. 17. (a) K
I
and (b) K
II
computed for different crack friction coefcient l
crack
(l = 6 mm, h = 15, l
wheelrail
= 0:025).
12 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
The SIF range DK
II
is thus dependent on the value of the crack friction coefcient. The crack growth rate and direction are
also modied as shown in Fig. 18. The crack propagation is divided into two periods. During the rst one, a co-planar crack
extension is obtained whatever the value of l
crack
. Then, differences in the crack path are predicted. Increasing l
crack
value
causes (i) an earlier crack branching and (ii) the crack branches downwards instead of upwards to the upper surface. This
is more dangerous as the former may lead to rail failure whereas the latter may simply lead to a spall.
5.3. Inuence of the initial crack orientation
The inuence of the crack inclination on the crack growth is analyzed in this section. The loading conditions and the crack
parameters are dened in Table 1. Different values of h are considered (15, 30, 45, 60 and 75) from shallow angled to
nearly perpendicular crack to the upper surface. K
I
and K
II
variations at the h angled crack tip are plotted versus the load
location during a cycle (see Fig. 19(a) and (b)). A shallow inclination reduces the relative slip between the crack faces but
eases the crack opening, for the congurations considered. Hence, DK
II
increases with increasing h while DK
I
decreases.
The crack growth driving force under mixed mode loading being a shear-tensile mechanism, the crack rate and branching
direction are inuenced. The crack branches downwards whatever h, except for h = 75 for which a co-planar extension is
predicted (see Fig. 20).
5.4. Inuence of the plastic stresses
The inuence of plastic stresses on the crack growth behavior is now investigated. The procedure described in Section 2.4
is used. This plastic stress eld has been computed on a different mesh using a dedicated numerical tool briey introduced in
Section 2.3. The plastic eld introduced is a combination of the residual stresses coming from the manufacturing process and
the ones deriving from the plastic strain accumulation due to the train trafc.
5.4.1. Plastic stress eld tensor
The components of plastic stresses r
plastic
xx
; r
plastic
yy
; r
plastic
xy
and r
plastic
zz
, in the center of the considered domain, are plotted
in Fig. 21. This gure shows that a complex stress state exists within the rail subsurface, with successive layers of different
thicknesses and different magnitude of tensile and compressive stresses: rst, tensile stresses (about 25 MPa) exist at the
surface of the rail, followed just below by high compressive stresses (about 110 MPa) within a 1 or 2 mm thick layer of
Fig. 18. Crack propagation path for different local friction coefcient l
crack
(l = 6 mm, h = 15, l
wheelrail
= 0:025).

15
30
45
60
75

15
30
45
60
75
6.00 4.00 2.00 0.00 2.00 4.00 6.00
15.00
20.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
15.00
20.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
6.00 4.00 2.00 0.00 2.00 4.00 6.00
Fig. 19. (a) K
I
and (b) K
II
computed different initial crack orientation h (l = 6 mm, l
crack
= 0:5; l
wheelrail
= 0:025).
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 13
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
the rail subsurface, then tensile stresses again (50 MPa) and nally again compressive stresses (maximum 50 MPa) for
10 mm depth. We also notice that shear stresses are localized just beneath the running surface and vanish in the rail
web. This plastic stress distribution results from the accumulation of plastic strain caused by repeated wheel rail contact
in the near surface vicinity and deeper in the rail subsurface from the manufacturing process, characterized by the C-shape
of residual stress eld. The r
xx
stress component is shown after projection on the mesh used for the simulation of the crack
propagation and the initial elastic re-balanced step (Fig. 21). As already mentioned, this initial state is considered as a non-
uniform and permanent state for the fracture problem.
5.4.2. Inuence of the plastic stress eld
SIF and the crack growth behavior are now analyzed and compared with and without accounting for the plastic stresses,
for the reference case (data in Table 1). The SIF are plotted in Fig. 22(a) and (b) for mode I and II respectively. As the plastic
stresses vary from tensile to compressive stresses from the crack mouth to the tip, they inhibit the crack opening during the
wheel traveling and K
I
= 0. DK
II
is further reduced as the compressive stresses decrease the relative sliding between the
crack faces.
Concerning the crack growth, the plastic stresses have a signicant inuence. First, the crack rate is reduced and sec-
ondly, after a longer co-planar extension, the crack branches slightly upwards to the free surface towards the zone sub-
mitted to tensile stresses (Fig. 23), instead of branching downwards. Two major consequences are highlighted through this
example: the plastic stresses may increase the fatigue life and may be responsible for a change in the damage mechanism,
from failure to spall for the considered case. Accounting for actual plastic stresses is thus of utmost importance for the
fatigue prediction.
The next step of this parametric study is concerned with multiple cracks and their interactions. In rails, cracks developped
often in networks, either on the rail head or at the gauge corner [40]. It is important to quantify the modications in the SIFs
Fig. 20. Crack propagation path versus the initial crack orientation h (l = 6 mm, l
crack
= 0:5; l
wheelrail
= 0:025).
Fig. 21. Plastic stress eld after projection on the mesh used for the propagation simulation on a vertical line at the center of the domain.
14 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
due to crack interactions with respect to the values obtained for a single crack and to expand this analysis to crack
propagation.
5.5. Crack network
The crack growth algorithm is modied in case of multiple cracks. The growth increment is imposed equal to Da
max
= 4%
of the current crack length in the single crack analysis. In the multiple crack analysis, the maximum growth increment is still
imposed equal to 4% and the crack growth increment for each crack i Da
i
is determined proportionally to the equivalent
stress intensity factor range to the power of the Paris law, such as:
4.00 2.00 6.00 0.00 2.00 4.00 6.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
4.00 2.00 6.00 0.00 2.00 4.00 6.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
FREE
WITH plastic stresses
FREE
WITH plastic stresses
Fig. 22. (a) K
I
and (b) K
II
computed with the plastic stresses (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:025).
Fig. 23. Crack propagation path with and without plastic stresses (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:025).
6 4 2 0 2 4 6
20.00
15.00
10.00
5.00
0.00
5.00
10.00
15.00
20.00
0.00
5.00
0.00
5.00
0.00
5.00
0.00
5.00
6 4 2 0 2 4 6
Fig. 24. (a) K
I
and (b) K
II
computed for a network of 2 cracks in a squat conguration (l = 6 mm, h = 15, l
crack
= 0:01; l
wheelrail
= 0:025). The reference case
corresponds to same conguration with only a single crack.
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 15
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
Da
i
Da
max
=
DK
eq
i
DK
eq
max
_ _
3:33
(32)
with
DK
eq
= DK
2
I
0:772 + DK
2
II
_ _
0:5
(33)
Fig. 25. Crack propagation path for 2 cracks in a squat conguration (l = 6 mm, h = 15, l
crack
= 0:01; l
wheelrail
= 0:025).
Fig. 26. Schematic representation of the initial conguration with 9 parallel cracks (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:1).
4 2 0 2 4 6
15.00
10.00
5.00
0.00
5.00
10.00
15.00
Fig. 27. (a) K
I
and (b) K
II
computed for a network of 9 parallel cracks (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:1). Reference stands for the SIFs obtained
for a single crack in the same conguration.
16 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
5.5.1. Two symetrical cracks
The rst case is concerned with two cracks labeled 1 and 2 (see Fig. 25). They have the same length 6 mm, their initial
inclination is h = 15, the crack friction coefcient is l
crack1
= l
crack2
= 0:5. The spacing is equal to 1 mm. Crack 1 surface
breaking tip is located at the origin of the reference axis. The load position is dened by x
c
=a. The loading conditions are
those dened in Table 1. K
I
and K
II
variations versus the load location are presented in Fig. 24 for cracks 1 and 2, accounting
for crack interactions. Two reference curves are added, corresponding to the SIF variations computed at crack 1 tip and crack
2 tip in the single crack conguration.
The effects of the interactions can thus be quantied. Differences in mode II SIF are not signicant. The major difference
concerns the crack opening of both cracks and thus DK
I
. Crack 1 opening is enhanced by the relative displacements along
crack 2 for x
c
=a < 2 while crack 2 opening is enhanced by the relative displacements along crack 1 for x
c
=a > 2.
The crack growth is then predicted. The results are presented in Fig. 25. Crack 1 branches and extends parallel to the rail
surface whereas crack 2 experiences a co-planar growth, as the one obtained in the one-crack conguration (Fig. 18). After
several propagations steps, DK
eq
1
is decreasing causing a reduction in the growth rate. These predictions are coherent with
observations made in [4143] for a similar crack conguration: the second crack propagates deeper in the rail than the rst
one.
5.5.2. Nine parallel cracks
The test case is presented in the Fig. 26. 9 initial parallel cracks, labeled from 1 to 9, of the same length (l = 6 mm, h = 15)
with equal spacing of 3:375 mm (equal to a=2, a quarter of the Hertzian contact zone) are considered. The crack coefcient of
friction l
crack
= 0:5 is the same for all cracks. l
wheelrail
= 0:1 (braking situation). Cracks labeled 3 to 6 are located at the center
of the network. K
I
and K
II
are plotted in Figs. 18(b) and 27(b) respectively, for the nine cracks. The reference SIF variations in
mode I and II, computed for the one-crack conguration, are also plotted for comparison and estimation of the interaction
effects. The inuence of the interactions between cracks are visible. Note that the crack inclination combined with the load
direction and the sign of the tangential load favors crack opening, enlarged for all cracks when compared to the one-crack
reference case, as the domain compliance is increased. A maximumDK
I
increase of 150% is obtained. Concerning the relative
sliding between the crack faces, the load sharing between the cracks leads to a shielding effect. DK
II
is thus signicantly
reduced (between 0 and 25 % depending on the crack).
The computations for the fatigue crack network propagation are then carried out. The fatigue crack growth is co-planar
before branching downward in the case of a single crack (Fig. 18(a)). At each load cycle, a growth increment is added to each
crack with respect to the value of the corresponding equivalent SIF range along the direction determined angle h
i
. The cal-
culation is rapidly stopped as the crack merging is not analyzed in this paper.
Fig. 28. Crack path after 5 propagation steps for a network of 9 parallel cracks h (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:1).
4 2 0 2 4 6
15.00
10.00
5.00
0.00
5.00
10.00
15.00
4 2 0 2 4 6
15.00
10.00
5.00
0.00
5.00
10.00
15.00
Fig. 29. (a) K
I
and (b) K
II
computed for a network of 9 parallel cracks with plastic stresses (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:1).
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 17
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
The crack path is represented for each crack of the network in Fig. 28. No signicant change is obtained for cracks 1, 2, 8
and 9 whereas cracks 3, 4, 5 and 7 branch parallel to the surface. The effects of the crack interactions are not the same for all
the cracks and are very complex. They depend for each crack on both the equivalent stress intensity factor range that governs
the crack rate and on the maximum equivalent stress intensity factor range at the tip of an innitesimal segment that deter-
mines the growth angle. It is interesting to notice that a shape of spalls is predicted for a few crack and propagation in the
depth of the rail for others.
5.5.3. Introduction of plastic stresses in the crack network
Finally, the simulation is carried out again but accounting for plastic stresses. Mode I and mode II SIFs are plotted in
Fig. 29(a) and (b) respectively. The compressive plastic stresses extend up to the crack tips and inhibit crack opening for
all cracks, except for crack 9 due to its particular location at the edge of the crack network. The relative sliding along all
the crack faces is also reduced and a signicant decrease in mode II stress intensity factor range is noted for all cracks.
The crack network propagation is also different. All cracks branch and extend in parrallel to the free surface in the direc-
tion of the moving load (see Fig. 30). Crack 9 behaves differently, being less inuenced by the crack interactions.
6. Conclusion
This paper aims at predicting fatigue crack growth and branching conditions under rolling contact fatigue taking into
account actual plastic stresses and not uniform ones. The model we present, combining the X-FEM discretization and the
LATIN nonlinear solver, is a robust tool capable of taking into account contact and friction at the interface and to simulate
efciently fatigue crack growth. Furthermore, the plastic elds are computed with a dedicated tool based on a direct station-
ary algorithm. Elastic shakedown is likely to occur for the stabilized rail mechanical state. Using this assumption and thanks
to the LATIN nonlinear solver, the introduction of palstic stresses is performed easily by only adding a generalized internal
force term in the right hand side of the equation describing the cracked body equilibrium. The developed numerical tool
within CAST3M [39] is formed by the proposed two-scale strategy taking into account the elds acting at the rail scale (plas-
tic stresses), the contact and friction between the crack faces at the crack scale, the fatigue Hourlier and Pineaus crack
branching criterion and Paris type fatigue crack growth laws. The X-FEM discretization of the structure problem allows very
easy propagation parametric studies since the crack mesh is not required to be conform with the structure one. Furthermore,
it is very interesting to keep the same mesh for a pamametric study. Indeed, all the simulations can then been compared at a
xed level of accuracy. The role of liquid is here assumed to simply reduce friction between the crack faces and to enhance
the relative movements.
The crack growth is inuenced by multiple parameters, in particular by plastic stresses, contact and interfacial crack fric-
tional effects and neighboring cracks. Hence, the variation of one of the mentioned parameters results in changes of the crack
propagation path. These parametric studies also give us a better understanding of the major damage observed under various
contact conditions. We have seen that the friction coefcient between the crack faces anf the wheel and the rail have an
antagonist effect on the growth rate. This point has been largely pointed out in the literature [5,7,8]. But in this paper prop-
agation simulation are led under all those conditions. This emphasizes an earlier crack branching when the friction coef-
cient between the wheel and the rail is high. Depending on the orientation of the tangential loading, rolling contact
fatigue cracks are expected to propagate upwards for accelerating train situations and downwards when the train is braking.
The work on the initial crack orientation has demonstrated that once cracks have branched downwards, for a given set of
parameters, a single crack follows the same propagation direction. This paper also demonstrates the inuence of plastic
stresses on the propagation path for fatigue cracks in rail. In this work we chose to place the crack tip in the zone of high
compressive stresses which lead to a decrease of the SIFs. The consequence was a longer coplanar propagation in the hor-
izontal strip of compressive stresses.
Those results rely on the study of a single crack in a 2D domain. The study of the squat conguration seems to show good
agreements with observations that were reported in some works [41,42]. The crack network investigation showed that crack
propagation behavior are modied and for the considering case that few cracks branch earlier in the horizontal direction
whereas others propagate downwards in the rail. A shielding effect has been observed for the K
II
values of the center cracks.
It is interesting to highlight that they are subjected to higher opening forces when dealing with a parallel crack network due
to the orientation of the tangential load and the crack initial inclination. The introduction of plastic stresses in the crack net-
work illustrates once again the high role of the plastic stress elds in the fatigue crack propagation behavior.
Fig. 30. Propagation path for a network of 9 parallel cracks without and with plastic stresses h (l = 6 mm, h = 15, l
crack
= 0:5; l
wheelrail
= 0:1).
18 B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020
Further improvements of the modeling include the rail bending or thermal loading and 3D crack propagation modeling.
The quantitative results require to be compared with the measurements coming from the railway network. This task implies
a statistical work to be able to compare the results of this deterministic model with the variations of very local effects in
practical situations.
References
[1] Bogdanski S, Lewicki P, Szymaniak M. Experimental and theoretical investigation of the phenomenon of lling the rcf crack with liquid. Wear
2005;258:12807.
[2] Bogdanski S, Lewicki P. 3D model of liquid entrapment mechanism for rolling contact fatigue cracks in rails. Wear 2008;265:135662.
[3] Farjoo M, Pal S, Daniel W, Meehan PA. Stress intensity factors around a 3d squat form crack and prediction of crack growth direction considering water
entrapment and elastic foundation. Engng Fract Mech 2012;94:3755.
[4] Fletcher D, Hyde P, Kapoor A. Modelling and full-scale trials to investigate uid pressurisation of rolling contact fatigue cracks. Wear
2008;265:131724.
[5] Fletcher D, Smith L, Kapoor A. Rail rolling contact fatigue dependence on friction, predicted using fracture mechanics with a three-dimensional
boundary element model. Engng Fract Mech 2009;76:261225.
[6] Farjoo M, Daniel W, Meehan PA. Modelling a squat form crack on a rail laid on an elastic foundation. Engng Fract Mech 2012;85:4758.
[7] Seo J, Kwon S, Jun H, Lee D. Fatigue crack growth behavior of surface crack in rails. Proc Engng 2010;2:86572.
[8] Dubourg M-C, Lamacq V. A predictive rolling contact fatigue crack growth model: onset of branching, direction, and growth: role of dry and lubricated
conditions on crack patterns. J Tribol 2002;124:6808.
[9] Brouzoulis J, Ekh M. Crack propagation in rails under rolling contact fatigue loading conditions based on material forces. Int J Fatigue 2012;45:98105.
[10] Ringsberg JW, Lindck T. Rolling contact fatigue analysis of rails inculding numerical simulations of the rail manufacturing process and repeated
wheel-rail contact loads. Int J Fatigue 2003;25(6):54758.
[11] Kapoor A, Fletcher DI. Post hateld rolling contact fatigue: the effect of residual stress on contact stress driven growth in rail. Part 1: The model. Tech.
rep., Newrail Report No. WR061106-2, Newcastle University; 2006.
[12] Kapoor A, Fletcher DI. Post hateld rolling contact fatigue: the effect of residual stress on contact stress driven growth in rail. Part 2: Data. Tech. rep.,
Newrail Report No. WR061106-3, Newcastle University; 2006.
[13] Kapoor A, Fletcher DI. Post hateld rolling contact fatigue: the effect of residual stress on contact stress driven growth in rail. Part 3: Further data. Tech.
rep., Newrail Report No. WR061106-4, Newcastle University; 2006.
[14] Kapoor A, Fletcher DI, Franklin FJ, Smith L, Hyde P. Post hateld rolling contact fatigue. the effect of residual stress on contact stress driven growth in
rail: comparison of the hateld and alternative uk rails using models to asess the effect of residual stress on crack growth from rolling contact fatigue.
Tech. rep., Newrail Report No. WR061106-6, Newcastle University; 2006.
[15] Seo JW, Goo BC, Choi JB, Kim JK. Effects of metal removal and residual stress on the contact fatigue life of railway wheels. Int J Fatigue 2008;30:20219.
[16] Bogdanski S, Brown MW. Modelling the three-dimensional behaviour of shallow rolling contact fatigue cracks in rails. Wear 2002;253:1725.
[17] Mellings S, Baynham J, Adey R. Automatic crack growth prediction in rails with bem. Engng Fract Mech 2005;72:30918.
[18] Nguyen-Tajan M, Funfschilling C. A numerical modeling strategy for the rolling contact fatigue analysis of rails. In: WCRR; 2011.
[19] Dolbow J, Mos N, Belytschko T. An extended nite element method for modelling cack growth with frictional contact. Comput Methods Appl Mecha
Engng 2001;190:682546.
[20] Ribeaucourt R, Baietto-Dubourg M-C, Gravouil A. A new fatigue frictional contact crack propagation model with the coupled x-fem/latin method.
Comput Methods Appl Mech Engng 2007;196:323047.
[21] Gniaut S, Massin P, Mos N. A stable 3d contact formulation for cracks using x-fem. Eur J Comput Mech 2007;16(2):25975.
[22] Liu F, Borja RI. Stabilized low-order nite elements for frictional contact with the extended nite element method. Comput Methods Appl Mech Engng
2010;199:245671.
[23] Pierrs E, Baietto M-C, Gravouil A, Morales-Espejel G. 3d two scale x-fem crack model with interfacial frictional contact: application to fretting fatigue.
Tribol Int 2010;43:183141.
[24] Gravouil A, Pierrs E, Baietto M-C. Stabilized global-local x-fem for 3d non-planar frictional crack using relevant meshes. Int J Numer Methods Engng
2011;88:144975.
[25] Mos N, Dolbow J, Belytschko T. A nite element method for crack growth without remeshing. Int J Numer Methods Engng 1999;46:13150.
[26] Melenk J, Babuska I. The partition of unity nite element method: basic theory and applications. Comput Methods Appl Mech Engng 1996;39:289314.
[27] Troll B, Gravouil A, Baietto M-C, Nguyen-Tajan TML. Optimization of a stabilized x-fem formulation for frictional cracks. Finite Elem Anal Des
2012;59:1827.
[28] Skyttebol A, Josefson B, Ringsberg J. Fatigue crack growth in a welded rail under the inuence of residual stresses. Engng Fract Mech
2005;72(2):27185.
[29] Dang Van K, Maitournam M, Moumni Z, Roger F. A comprehensive approach for modeling fatigue and fracture of rails. Engng Fract Mech
2009;76:262636.
[30] Nguyen Q. On the elastic plastic initial-value problem and its numerical integration. Int J Numer Methods Engng 1977;11:81732.
[31] Dang Van K, Maitournam M. Steady-state ow in classical elastoplasticity: application to repeated rolling and sliding contact. J Mech Phys Solids
1993;41(11):1691710.
[32] Bold PE, Brown M, Allen R. A review of fatigue crack growth in steels under mixed-mode i and ii loading. Fatigue Fract Engng Mater Struct
1992;15:96577.
[33] Hourlier F, Dhondt H, Truchon M, Pineau A. Etude du branchement des ssures de fatigue sous sollicitations bimodales non proportionnnelles:
inuence de la cintique de propagation du matriau. Tech. rep., Institut de reccherche de la sidrurgie franaise IRSID; 1982.
[34] Lamacq V, Dubourg M-C. Modelling of initial fatigue crack growth and crack branching conditions under fretting conditions. Fatigue Fract Engng Mater
Struct 1999;22:53542.
[35] Dubourg M-C, Lamacq V. Stage ii crack propagation direction under fretting fatigue loading: a new approach in accordance with experimental
observations. In: Hoeppner DW, Chandrasekaran V, Elliot CB, editors. Fretting fatigue: current technology and practices, ASTM, 1367, American Society
for Testing and Materials, Baltimore, MD; 2000. p. 43650.
[36] Baietto M-C, Pierrs E, Gravouil A, Berthel B, Fouvry S, Troll B. Fretting fatigue crack growth simulation based on a combined experimental and xfem
strategy. Int J Fatigue 2013;47:3143.
[37] Dubourg M-C, Villechaise B. Analysis of multiple fatigue cracks. Part i: Theory. ASME J Tribol 1992;114:45561.
[38] Dubourg M-C, Godet M, Villechaise B. Analysis of multiple fatigue cracks. Part ii: Results. ASME J Tribol 1992;114:4628.
[39] http://www cast3m.cea.fr/.
[40] Tillberg J, Larsson F, Runesson K. A study of multiple crack interaction at rolling contact fatigue loading of rails. Proc Inst Mech Eng, Part F: J Rail Rapid
Transit 2009;223(4):31930.
[41] Cannon D, Pradier H. Rail rolling contact fatigue research by the european rail research institute. Wear 1996;191:113.
[42] Steenbergen M, Dollevoet R. On the mechanism of squat formation on train rails. Part i: Origination. Int J Fatigue 2013;47:36172.
[43] Grassie SL, Fletcher DI, Hernandez EAG, Summers P. Studs: a squat-type defect in rails. Proc Inst Mech Eng, Part F: J Rail Rapid Transit 2012:24356.
B. Troll et al. / Engineering Fracture Mechanics xxx (2014) xxxxxx 19
Please cite this article in press as: Troll B et al. 2D fatigue crack propagation in rails taking into account actual plastic stresses. Engng Fract
Mech (2014), http://dx.doi.org/10.1016/j.engfracmech.2014.03.020